I Built an AI to Nag Me About My Health—Here's How You Can Too

Let’s be honest, I’m terrible at sticking to health goals. The world is full of well-meaning apps and smartwatches that beep and buzz, reminding me to drink water, stand up, or breathe. And I have become exceptionally good at ignoring every single one of them. The notification to drink a glass of water might as well be an ad for a product I’ll never buy. It’s impersonal, generic, and easy to swipe away into the digital void.

My primary challenge is hydration. I live in Mumbai, where you can lose a litre of water just by thinking about walking outside. Yet, I consistently forget to drink water. I’m a coffee-and-chai person. By 4 PM, I’m running on caffeine and fumes, nursing a mild headache, and wondering why. My Fitbit buzzes, my phone flashes, and I dismiss them all. They don’t understand my excuses. They don’t get that the “critical meeting” is often just me doomscrolling on Twitter, or that my “just one more chai” is a sacred ritual.

I needed something different. I didn't need another notification; I needed a nudge. A personalized, persistent, and slightly annoying nudge. I needed something that knew my goals, understood my pathetic excuses, and could call me out in a language I would actually listen to. I didn’t need an app. I needed an AI aunty.

This is where my journey into Custom GPTs began. For those who aren't deeply embedded in the AI world, the term might sound intimidating. It sounds like something that requires a degree in computer science and a fluency in Python. I’m here to tell you it doesn’t. If you can write an email or a text message, you can build your own Custom GPT. In essence, a Custom GPT is just ChatGPT with a specific job description. You give it a name, a personality, a set of instructions, and maybe some reference documents. It then transforms from a generalist know-it-all into a specialized assistant designed by you, for you.

My goal was simple: to build a “Health Nagger” AI. I wanted it to be my personal, digital drill sergeant, but with the distinct flavour of a loving but perpetually disappointed Indian aunty. Someone who would say, “Beta, did you drink water?” but would also follow up my excuse of being “busy” with, “Busy? Are you running the country? Go drink some water before you shrivel up like a raisin.”

The process was surprisingly simple and, dare I say, fun. It took me less than an hour, most of which was spent laughing at my own creation.

First, I opened up ChatGPT (you need a Plus subscription for this) and navigated to the ‘Explore GPTs’ section. There, a button beckoned: ‘Create a GPT’. Clicking it opened a new interface, split into two parts. On the left was the ‘Create’ panel, which is a chat window where you talk to a ‘GPT Builder’ to set up your new AI. On the right was the ‘Preview’ panel, a sandbox where you can test your creation in real-time. It’s like teaching a new employee a task and then immediately watching them try it out.

My conversation with the GPT Builder began. “I want to create a GPT that nags me to stick to my health goals,” I typed.

The builder responded instantly, asking for more details. What should we call it? I decided on “Wellness Aunty.” The builder then suggested a profile picture, generating a pleasant-looking woman with a warm but firm expression. Perfect.

Next, it asked about the goals. I listed them out: Drink 3 litres of water a day. Walk 8,000 steps. Stretch for 5 minutes every hour I’m at my desk. Avoid fried snacks after 6 PM (a particularly difficult one in a city famous for its vada pav).

The crucial part was defining the personality. I wrote: “I want it to have the personality of a caring but very strict Indian aunty. It should be a bit sarcastic and witty. It should not accept my excuses easily. It can use endearing terms like ‘beta’ but also be very direct and call me out on my laziness. It should feel like it knows me.”

The builder processed this and updated the GPT’s core instructions. Then came the fine-tuning. The builder asked how Wellness Aunty should handle specific scenarios. What if I say I'm too tired to walk? What if I say I had a stressful day and deserve a samosa? For each, I gave it a sample of the kind of response I wanted. “If I say I’m tired,” I instructed, “it should say something like, ‘Tired from all that sitting? A short walk will give you more energy than that third cup of coffee, you know.’”

This back-and-forth is the magic of the creation process. You’re not filling out a form; you’re having a conversation. You’re teaching and shaping the AI’s personality through simple dialogue.

After this initial chat, I switched to the ‘Configure’ tab. This is where you can see and directly edit the instructions the builder has written based on your conversation. It’s the behind-the-scenes look. My instructions looked something like this:

“You are Wellness Aunty. Your goal is to help Rohan stay on track with his health goals. You are to be firm, a little sarcastic, and use the tone of a loving but strict Indian aunty. His goals are: [Goals listed here]. When he makes an excuse, you must challenge it smartly. For example, if he says he’s in a meeting, ask him if he can’t drink water during a meeting. Remind him of his long-term goals.”

I even added a line: “You are aware of Indian cultural contexts, like the temptation of evening chai and snacks.” This small detail makes a world of difference. A generic AI might not understand the gravitational pull of a hot plate of bhajiyas on a rainy evening. Mine does.

One of the most powerful features is the ability to upload ‘Knowledge’. I downloaded a simple one-page PDF of desk stretches and uploaded it. Now, when Wellness Aunty tells me to stretch and I ask, “What stretches should I do?”, it can pull information directly from that document. It’s not just nagging me; it’s providing the solution right there. It has its own little library that I gave it.

With the setup complete, it was time to test it in the Preview panel. I started a conversation with Wellness Aunty.

Me: “Hi Aunty-ji, I’m feeling very sluggish today.”

Wellness Aunty: “Hello beta. Sluggish, you say? Let me guess, you have had more coffee than water so far. Am I right? Go on, have a big glass of water right now and then tell me how you feel.”

It was perfect. The tone was spot on. It had the hint of accusation, the actionable advice, and the affectionate ‘beta’ all rolled into one.

I decided to push it. “I’m too busy with work for a walk today.”

Wellness Aunty: “Busy? I see. Was it you I saw scrolling through Instagram for 20 minutes a little while ago? Your legs work even when your brain is tired. A 15-minute walk around the block, beta. No excuses. The report can wait.”

I was floored. And a little bit called out. This was not a notification I could just swipe away. This was a conversation. The personalization made it feel compelling. It felt like someone who knew me was holding me accountable. It was effective precisely because I had designed it to be.

In the weeks since, Wellness Aunty has become a fixture in my daily routine. I have the ChatGPT app on my phone, and I check in with it a few times a day. It doesn't magically make me a health icon, but my water intake has definitely gone up. The simple act of having to articulate my excuses to an AI that I know will dismantle them often makes me just do the healthy thing instead.

The implications of this go far beyond nagging me about water. You could create a Socratic tutor to help you learn a new subject, a brainstorming partner that knows your company’s brand voice, or a travel planner that understands your exact preferences for quiet, non-touristy spots. The power is in the specificity.

What this experience taught me is that the next phase of our relationship with AI won’t just be about using monolithic, one-size-fits-all models. It will be about creating small, bespoke AIs that serve our individual needs. It’s a shift from being a passive consumer of technology to being an active creator of personal tools.

Building Wellness Aunty demystified AI for me. It showed me that this powerful technology isn’t some arcane magic accessible only to engineers in Silicon Valley or Bengaluru. It’s a tool, like a word processor or a spreadsheet, that is becoming increasingly malleable and accessible. All it takes is a clear idea and a little bit of conversation to shape it into something genuinely useful for your own life. Now, if you’ll excuse me, Aunty-ji is telling me I’ve been sitting for too long.